The Impact of News Consumption on Anti-immigration Attitudes and Populist Party Support in a Changing Media Ecology

ABSTRACT Democracies around the world are facing a rising wave of right-wing populism and new nationalism, which often relies on strategic exploitation of anti-immigration sentiments. While media have long been acknowledged as important channels of anti-immigration rhetoric, the evidence of the actual impact of news consumption on attitudes to migration and support for populist parties is still inconclusive, and largely limited to pre-digital media ecologies. Combining a representative two-wave panel survey (N=819), digital tracking of real-time electronic measurement of television, radio and online media exposure, and an analysis of news content, this study explores the effect of news consumption on anti-immigration attitudes and electoral behaviour during the EP2019 election campaign in the Czech Republic. Our analysis reveals that being exposed to news about migration – particularly on websites and on commercial television stations – increases the likelihood of voting for populist parties, while exposure to public service media leads to less negative attitudes towards immigration. At the same time, being exposed to more news sources intensifies, rather than reduces, anti-immigrant attitudes. This result challenges the assumption that a more diverse news media diet could serve as an antidote to selective exposure and “echo chambers”, commonly linked with radicalization of political views.


Introduction
In many Western countries, the issue of immigration has been one of the most debated and polarizing topics for several years. 1 In Europe, concerns surrounding immigration became particularly prominent after the sharp rise in immigration from war-torn Syria in 2015 and remain high to this day (European Union, 2018). Several right-wing populist parties in Europe, from the Alternative for Germany and Marine Le Pen's National Rally to Matteo Salvini's Lega Nord, have (re)gained prominence and electoral support on the back of the migration crisis. Strategic exploitation of anti-immigration rhetoric also played a key role in the outcome of the Brexit referendum in the UK and the 2016 U.S. Presidential election; the "Vote Leave" campaign and Donald Trump's "Make America Great Again" campaign both mobilized voters by appealing to anxieties surrounding immigration (Goodwin & Milazzo, 2017;Hooghe & Dassonneville, 2018).
The media are often acknowledged as an important factor that can help explain the rise in anti-immigration sentiments and public support for right-wing populism (e.g., Boomgaarden & Vliegenthart, 2007;Bos et al., 2019;Van Klingeren et al., 2015;Wirz et al., 2018). A wealth of research has examined how anti-immigration discourse is constructed and perpetuated in press and broadcast media, showing how immigration is consistently represented and framed in negative terms, associated with criminality or terrorism, or presented as a threat to cultural identity and an economic burden (e.g., Eberl et al., 2018;Lawlor & Tolley, 2017;Georgiou & Zaborowski, 2017). Yet, while the mechanisms of anti-immigration discourse in the media are by now well known, the evidence of impact of such discourse on audiences is far less conclusive. First, existing research often relies on cross-sectional survey designs or experiments and hence offers either evidence of correlation rather than causation, or evidence of causation without sufficient external validity. Second, information about people's media consumption habits is usually derived from questionnaires, and therefore relies on self-reported media exposure rather than actual media exposure. Third, the majority of existing research is limited to long-established democracies of Western Europe and Northern America. And finally, little is known about how anti-immigration attitudes and right-wing support are shaped in a changing media ecology, one that-despite the continuing importance of television as a news source (Newman, 2019)-is characterized by the rise of digital platforms and by the establishment of a hybrid media system (Chadwick, 2013). Right-wing populist actors have eagerly adopted the new digital communication channels, bypassing the gatekeeping role of press and broadcasting to reach the public directly Ernst et al., 2017;Krämer, 2017;Pajnik & Sauer, 2017). However, what exactly this means for citizens and their vulnerability to xenophobic and anti-immigration messages remains an open question.
In an attempt to fill these multiple research gaps, this study adopts a mixed-method approach combining a two-wave panel survey, digital tracking of real-time cross-media consumption, and analysis of immigration-related news media content to gauge the effect of news exposure on attitudes toward immigrants and support for populist parties. The study was conducted during the 2019 European Parliament election campaign in the Czech Republic, a Euroskeptic country currently led by a populist government and marked by high levels of anti-immigration sentiment.

News Media, Anti-immigration Attitudes, and Support for Right-wing Populism
The relationship between media and migration, and specifically the role of the media in fostering anti-immigrant attitudes and fueling the rise of right-wing populist parties and leaders, has become an increasingly studied subject over the last decade (Eberl et al., 2018, p. 208). This trend is a reflection of the changing political and social realities, including the growing importance of immigration in public debate, and the use of anti-immigration discourse among right-wing populist parties and leaders, who rely on exclusionary rhetoric to garner public support (Matthes & Schmuck, 2017, p. 556).
It is therefore not a surprise that researchers have become more attentive to the effects of (negative) immigration coverage; while public attitudes to immigration and public support for right-wing parties have traditionally been explained primarily with reference to demographic, economic and political factors-such as immigration rates, size of the immigrant population, unemployment rates, and events such as terrorist attacks-the media now often feature as one of the main explanatory variables (e.g., Bos et al., 2019;Boomgaarden & Vliegenthart, 2007;Van Klingeren et al., 2015;Wirz et al., 2018).
Theoretically, the majority of existing research in this area approaches the research problem from the perspective of agenda setting or framing research traditions (for an overview see Eberl et al., 2018), typically combined with social-psychological theories of social identity and inter-group conflict (e.g., Tajfel & Turner, 1986). In line with these theoretical frameworks, this body of work examines the impact of exposure to immigration-related media content, typically using either a combination of survey data and quantitative media content analysis, or experimental methodologies. Two dimensions of immigration-related content tend to be at the forefront: salience and valence.
One of the earliest studies examining the impact of salience of immigration news, based on a combination of content analysis of Belgian newspapers and television news, and election polling data from Belgium between 1991 and 2000, found that immigration and crime received growing media attention that paralleled the electoral growth of the right-wing Vlaams Blok (Flemish Block), leading the authors to conclude that "the media could be considered coresponsible for the Vlaams Blok's upsurge" (Walgrave & De Swert, 2004, p. 479). However, a study conducted in Germany between 1993 and 2005 found that the amount of immigration news was only marginally positively related with negative perceptions of immigration (Boomgaarden & Vliegenthardt, 2009). Research by Van Klingeren et al. (2015) found that the salience of immigration coverage does have an impact on attitudes towards immigration and tends to increase negative attitudes, but that this impact can vary from country to country. Finally, a comparative panel study of voters in 11 European countries demonstrated that exposure to news about immigration (as well as about crime) increases chances of people casting a vote for an anti-immigration party (Burscher et al., 2015).
In contrast to existing research on the impact of salience of immigration news, the results of studies of the impact of valence are more consistent (Boomgaarden & Vliegenthardt, 2009;Dixon & Azocar, 2007;Schemer, 2012;Van Klingeren et al., 2015;Wirz et al., 2018). Boomgaarden and Vliegenthardt (2009) found that positive coverage of immigration was associated with lower levels of public concern about immigration, while negative coverage was associated with higher levels of public concern. A similar conclusion about the impact of immigration news valence was reached by a study of effects of immigration news during a political campaign concerning naturalization procedures for immigrants in Switzerland in 2008 (Schemer, 2012), showing that negative news coverage of immigrants increased stereotypical attitudes, while exposure to positive portrayals reduced them. These conclusions received further support by a four-country study that likewise concluded that "the more anti-immigrant messages individuals received in their media diets, the more negative their cognitions became over time" (Wirz et al., 2018, p. 507). Finally, in the Netherlands, Sheets et al. (2015) did not find a direct effect of antiimmigrant newspaper messages on the vote for a populist party, but their experimental study revealed an impact on anti-immigrant attitudes (Sheets et al., 2015).

Different Media Types, Different Effects?
While there has been growing evidence of the general effects of anti-immigration news content on anti-immigrant attitudes and/or electoral support of right-wing populist parties and actors (Hameleers, 2019), less is known about the impact of different types of media. For instance, is reliance on commercial and tabloid media likely to strengthen anti-immigrant attitudes? Can strong public service media help foster more positive attitudes to immigration? And finally, what are the consequences of the growing reliance on digital news?
Scholarship on populism has long maintained the existence of a correlation between consumption of tabloid and other "soft"/entertainment media, and support for populist parties, usually explained by the affinity of these media with the populist communication style Krämer, 2014;Mazzoleni, 2008;Schulz, 2019). However, there has been less clarity about the direction of causality in this relationship. A recently published study (Durante et al., 2019) claims to have proven long-term impact of early exposure to commercial TV in Italy on the likelihood of voting for the party of Silvio Berlusconi-the owner of the network-in the 1994 Italian general election. Studies carried out in Belgium and the Netherlands have suggested that exposure to commercial television is positively related to negative perceptions of immigrants and/or ethnic minorities (Jacobs et al., 2016;Vergeer & Scheepers, 1998). On the other hand, exposure to public service broadcasting was linked to more positive attitudes to immigration than exposure to commercial TV in some countries (Beyer & Matthes, 2015). One of the largest multi-national studies so far on the subject (Schulz, 2019) likewise found an association between strong populist attitudes and the frequency of use of tabloid press as well as commercial TV.
How about the impact of digital media? At first sight, general arguments about the nature of the new information ecology seem to suggest that greater reliance on digital news is likely to strengthen anti-immigration sentiments for several reasons. The participatory nature of digital media, combined with the lack of editorial oversight and fact-checking procedures, has made it easier for individuals and groups to spread extremist views and misinformation (Nichols, 2017;Rojecki & Meraz, 2016)-an argument that can be applied also to the spreading of antiimmigration discourse. At the same time, digital media-and social networking platforms in particular-have been blamed for encouraging the fragmentation of public debate by means of creating algorithm-driven "filter bubbles" (Pariser, 2011) or "echo chambers" (Garrett, 2009;Sunstein, 2018) within which people are selectively exposed to views that conform to their existing political attitudes and beliefs. Again, it is easy to see how such selective exposure may work to strengthen anti-immigration sentiments among audiences who are already negatively inclined toward the issue. At the same time, selective exposure may also be strengthening the pro-migration stance among those who adopt a more welcoming stance to immigration, thus resulting in a growing polarization of public opinion on this issue. That said, much depends on the nature of content available through social media, specifically its valence. In a digital media environment dominated by anti-immigration views, potential for polarization may be rather limited, with most social media users slowly becoming more negative toward immigration. In sum, these arguments suggest that the changing information ecology may be a contributing factor in the proliferation of anti-immigrant attitudes and in the rise of right-wing populism, while also potentially fostering polarization.
Yet, empirical evidence of the impact of the changing information environment is far from unanimous (for overviews see Tucker et al., 2018;Van Aelst et al., 2017). Alarmist claims about the extent and effects of echo chambers have been questioned as well, with one systematic literature review published in 2016 concluding that "at present, there is little empirical evidence that warrants any worries about filter bubbles" (Zuiderveen Borgesius et al., 2016). For instance, based on a representative sample of UK Internet users, Dubois and Blank (2018) found that interest in politics and a diverse media diet tend to limit the extent to which people are caught in echo chambers, while a recent review of evidence conducted by Guess et al. (2018) suggests that online media use increases exposure to diverse news sources, and vulnerability to echo chambers may in fact be greatest in off-line information environments. Worth noting here is the underlying assumption that diversity of sources is a good thing-and conversely, that selective exposure to a limited range of information sources is more likely to mean that media users are caught in an echo chamber and hence more likely to develop more extreme views. To sum up, regardless of whether or not the new information ecology leads to the formation of echo chambers, there is an agreement that such echo chambers can be counteracted through exposure to diverse news sources. By analogy, one could expect that those consuming a more diverse media diet will be less likely to embrace anti-immigrant rhetoric.
Available empirical research that tackles the impact of the changing information environment and echo chambers on anti-immigrant attitudes and support for right-wing populism is scarce. Of most direct relevance to our research are the results of the previously mentioned multi-national study that examined the link between populist attitudes and news consumption across both "old" and "new" media (Schulz, 2019). While the study found no difference among the "digital-only" and traditional media users, it showed that people exhibiting stronger populist attitudes preferred Facebook as a source of political information, while those with lower populist attitudes preferred Twitter. Also important for our investigation is a six-country study that examines the links between populist attitudes and online news consumption (Stier et al., 2020). In line with existing research on the selective exposure framework, this study shows that populist attitudes correlate with higher reliance on hyper-partisan online outlets and lower reliance on legacy press websites, but it also shows that populist citizens do not avoid public service news, rely overwhelmingly on legacy press websites rather than hyper partisan sources, and-surprisingly perhaps-consume less tabloid news (ibid., p. 13). Cross-country differences are also notable, suggesting that the link between populist attitudes and news diets is to an important extent shaped by the nature of the online media ecology in each country.
We should also note the body of research that focuses specifically on the impact of antiimmigrant discourse online, using online experiments (Bos et al., 2019;Hameleers & Schmuck, 2017;Hameleers., 2019). While the exclusive focus on online media makes it difficult to know how online anti-immigrant discourse might function in a complex, hybrid media environment, some key conclusions of this body of work are worth noting. Interestingly, all existing studies found evidence of significant disparities in impact of antiimmigrant discourse across different groups of participants. A comparative experiment conducted in Austria and the Netherlands found that messages blaming the Elites or immigrants strengthened populist attitudes, but only among those who supported the source of the message, while leading to the opposite effect among those who opposed it (Hameleers & Schmuck, 2017). Another experimental study demonstrated that messages that blamed immigrants for crime led to negative stereotypes of immigrants, but only among people with stronger perceptions of relative deprivation, while having the opposite effects among citizens with weaker perceptions of relative deprivation (Hameleers., 2019).

Research Questions and Hypotheses
Despite expanding scholarship in the area of news consumption and electoral support for anti-immigration parties, there are clearly numerous research gaps that prevent us from drawing a more comprehensive picture of this relationship-from fragmented and sometimes contradictory evidence concerning exposure to different media types, to limited crossnational comparability of the findings. The validity of existing findings can be questioned as well, as existing research is mostly-with one notable exception (Stier et al., 2020)-based on self-reported (as opposed to actual) media use and exposure to, which is known to suffer from systematic misreporting and exaggeration .
In order to advance the understanding of the effects of anti-immigration discourse in an increasingly complex, interconnected media environment, this study employs an innovative research design that (a) combines multiple data sets; (b) adopts a longitudinal rather than cross-sectional approach, enabling researchers to measure the effect of news consumption; and (c) utilizes a reliable measure of actual exposure based on digital tracking (as opposed to exposure that relies on self-reporting).
The study is driven by the following research questions:

RQ1: How have people's media repertoires affected people's attitudes toward immigration and electoral behavior during the 2019 EP election campaign?
Uncoupling the complex patterns of relationships implied by this question, we first wanted to examine the impact of specific types of media and the impact of the diversity of news sources:

RQ1.2: Have people's attitudes toward immigration and their voting choices been affected by the diversity of media they used during the campaign?
Based on evidence which suggests that a more diverse media diet decreases exposure to echo chambers (Dubois and Blank, 2018) and evidence suggesting that "people exposed to more than one newspaper perceive ethnic minorities as less threatening than others do" (Vergeer et al., 2000, p. 140), we predict: H1: People with less diverse media diets during the campaign will be more likely to adopt more negative views on immigration and vote for populist parties.
Furthermore, we are interested in whether there is a difference between exposure to commercial television and public service television in the change of migration-related attitudes and electoral choices (RQ1.3), and we expect-based on findings by Vergeer and Scheepers (1998), Jacobs et al. (2016), or Beyer and Matthes (2015)-that: H2: More intense consumption of commercial television, as opposed to public service television, will lead to more negative attitudes toward immigrants and higher likelihood of voting for populist parties.
Apart from exploring the general impact of media exposure, we are also interested in the effect of consumption of actual news about immigration:

RQ2: How has the consumption of news about immigration affected people's attitudes and electoral behavior during the 2019 EP campaign?
Building on the scholarship suggesting a link between salience of immigration-related news and anti-immigration attitudes and populist support (Burscher et al., 2015;Van Klingeren et al., 2015), and following the previously quoted studies that point to the amplifying effect of commercial television on anti-immigration attitudes, we expect that: H3.1: Heavier consumers of news about immigration in general will become more negative toward immigration over the course of the campaign, and (H3.2) they will be more likely to vote for parties with anti-immigration agenda; as well as H3.3: Heavier consumers of news about immigration that will appear on commercial television channels, as opposed to public service media, will become more negative toward immigration over the course of the campaign and (H3.4) they will be more likely to vote for parties with anti-immigration agenda.
Finally, drawing on the studies accounting for the effects of valence of immigration news (Boomgaarden & Vliegenthardt, 2009;Schemer, 2012;Wirz et al., 2018), we expect that (H3.5) exposure to negative news on immigration will lead to an increase of anti-immigration attitudes and (H3.6) to increased likelihood of voting for populist parties.

Case Study
The study was conducted during the 2019 European Parliament elections and focused on the Czech Republic, which constitutes a particularly intriguing case for exploring the impact of news exposure on anti-immigration attitudes and populist support. The Czech Republic has been largely unaffected by the 2015 migration crisis, corresponding with a very low number of asylum seekers and comparatively low percentage of migrants and foreigners in general. 2 Despite this, several right-wing populist political movements and parties have successfully exploited anti-immigration attitudes of the Czech population over the last several years.
With regards to patterns of news consumption, online platforms (websites and social media) are the most prevalent sources of news in the Czech Republic according to the 2019 Reuters Institute's Digital News Survey (2019), with 85% of respondents using them weekly, followed by television (77%) (Newman et al., 2019). However, the news aired by the main national TV stations have still the largest audience reach, led by the public service Czech TV (56% weekly usage), followed by the two biggest commercial TV stations (TV Nova, 50%, and TV Prima, 43%). The public service Czech Radio is the most popular radio news provider (15%), while the most accessed news websites reach between 30-40% of online news consumers. Facebook is by far the most dominant social media platform, being used by 75% of the Czech Internet population, 50% using it for news as well (Newman et al., 2019).
The 2019 EP elections brought to the polls 28.7% of eligible Czech voters, which was 10% more than in 2014. The election campaign was rather low-key and mostly devoid of any tangible policy issues campaigned for by the parties. Among the few exceptions was the issue of immigration, which was used by several parties to mobilize voters , including the two parties usually labeled as populist, the Prime Minister Babiš's party ANO, and the radical right-wing Freedom and Direct Democracy party (SPD). ANO's main slogan for the election was "We will protect Czechia. Strictly and adamantly," which contained a thinly veiled reference to the migration crisis, while SPD resorted to more open forms of anti-immigration rhetoric, accompanied with an adaptation of Donald Trump's signature slogan "Czech Republic First!" (Štětka, 2019). The immigration card brought, however, rather unimpressive results for the two parties, as ANO-despite overall victory-received only 21.2% of votes (much less than expected), while SPD was also aiming higher than the 9.1% of votes with which it ended up.

Data and Measurement
Data were collected in the Czech Republic between 26 April and 28 May 2019, before, during, and after the European Parliamentary elections (24-25 May). The quantitative analysis uses three datasets.
First, a panel survey, comprised of two questionnaires, was administered prior to the "hot phase" of the election campaign (three weeks before the elections) and immediately following the elections (four days). The surveys were conducted by Median, a Czech research agency, combining a CAWI + CATI method of data collection and utilizing the company's own panel for the measurement of cross-media consumption (see below), composed to be representative for the Czech population aged 15-69 on the key sociodemographic quotas (age, gender, education, region and size of municipality; the quotas have been constructed based on data from the Czech Statistical Office). Questionnaires were sent to 1180 panelists (18+); they were completed by 938 respondents in the first wave and by 819 respondents in the second wave (attrition rate 13%). A total of 97% of questionnaires were distributed online (CAWI), the rest via telephone interviews (CATI). There was a slight overrepresentation of women in the sample (55%) and of younger age cohorts (median age was 44, compared to 47 for the Czech 18+ population). In terms of education, 9% of respondents had an elementary education, 31% secondary education without a diploma, 42% secondary education with diploma, and 18% had a university education, which is broadly consistent with the proportions in the general population 24-64 (6-34-37-21% in 2018 according to the Czech Statistical Office; % for 18+ unavailable). The surveys were used to measure our dependent variables (attitudes toward immigration and electoral behavior/preferences).
Second, data on media consumption were collected on the same panel (the 819 respondents that took part in both waves of the survey) using adMeter, an innovative, state-of-the-art digital data tracking technology that enables real-time electronic measurement on people's media consumption across various media types. AdMeter is a mobile phone application (synchronized with other devices such as PC, laptop, or tablet) that, apart from tracking consumption of online sources (websites and social media), constantly listens to its surroundings and sends the recorded data (in an encrypted format) to a server that is simultaneously processing live streaming of the major Czech television and radio stations. The two data sources (from adMeter and from radio/TV streaming) are subsequently matched and analyzed by the computer which then determines whether or not the respondent was exposed to the particular channel. 3 The dataset thus provides information on the source, type, and duration of media that each survey participant has been exposed to, allowing it to measure the impact of that exposure on individual attitudes.
The third dataset consists of migration-related news items that were aired or published by key media outlets during the campaign period (see Appendix A for their full overview). This dataset was collected by the Czech branch of Newton Media, an international media monitoring and analysis company. Specifically, television, radio, and news website consumption was recorded whenever selected migration-related keywords were identified in the headlines or in the lead paragraph (or, in case of TV/radio reports, among the first 100 words in the transcript): "migration"/"immigration," "migrants"/"immigrants," and "refugees" (in their respective Czech versions, see Appendix A). Matched with the adMeter data using anonymized panelist ID, this dataset-comprising 267 news items from 40 news sources-thus facilitates further analysis of the effects of media exposure by allowing identification of specific, migration-related content in news sources, as well as the amount of time that respondents were exposed to such media. In addition, an analysis of valence was carried out on all these 267 items to determine whether their coverage of immigration was positive, neutral, or negative. 4 All respondents gave an explicit consent to their data being used for the study purposes, and the study received an ethical clearance by Loughborough University on 21 March 2019.

Dependent Variables
To test the hypotheses of the paper, the analysis examines four dependent variables focusing on popular attitudes and voting patterns regarding immigration and populist support in the Czech Republic.
The first dependent variable is an index measuring individuals' anti-immigration attitudes. This is made up of four items (inspired by Van Elsas et al., 2019), measuring agreement on an 7-point Likert-scale scale with the statements: (a) "immigrants abuse the welfare system"; (b) "immigrants are a threat to our culture"; (c) "immigrants cause the rise of criminality"; and (d) "immigrants have a positive effect on the labour market." The variable, constructed as a sum of the individual scores (with the last item reverse coded), has a 25-point range, where higher values denote stronger anti-immigration attitudes. Mokken scale analysis shows that the four items have a strong scalability of Loevinger's H coefficient of 0.58.
The second is a binary outcome variable measuring whether the respondent voted for a populist party; i.e., ANO (Action of Dissatisfied Citizens) and SPD (Freedom and Direct Democracy).
The third and fourth dependent variables examine how individuals' voting intention for ANO and SPD has changed from the pre-to the post-campaign period. In the pretest measurement, the variable combines three questions, yielding a 4-point scale that assigns a value of 3 to those who prefer ANO/SPD and are definitely decided prior to elections; 2 if they prefer ANO/SPD and are almost decided; 1 if they prefer ANO/SPD but are not yet decided OR do not prefer ANO/SPD but would consider voting for them; and 0 if they would not even consider voting for ANO/SPD. Similarly, the posttest variable is composed of two variables, but measuring vote in the EP elections and identification with the party. Accordingly, the 4-point scale gives a value of 3 to those who voted for ANO/SPD and strongly identify with it; 2 if they voted for ANO/SPD and have a neutral opinion on it; 1 if they voted for ANO/SPD and weakly identify with it; and 0 if they did not vote ANO/SPD.

Independent Variables
The first set of independent variables is concerned with the effect that intensity and diversity of media have on the outcome variables. Intensity of media use measures total time of exposure (hours) to a particular media source, whilst diversity of media sources counts how many different sources the respondent has been exposed to. Diversity of media sources has been operationalized in two ways. First, we measured the total number of sources a respondent has been exposed to. Second, we measured the number of sources a respondent was exposed to by the type of media technology: television, radio, news websites, and social media (in hours). Two additional variables focus specifically on the intensity of exposure to public service media and commercial television (in hours).
A second set of independent variables examines respondents' exposure to immigration coverage; i.e., content that contained at least one of four immigration keywords. These measurements are operationalized in two ways: total exposure in hours (continuous variable) and whether the respondent has been exposed to a particular type of media that contained those keywords or not (a binary variable). For each of these, they are applied to, first, general exposure to any media containing the immigration keywords, and second, a set of variables measuring exposure to immigration content through different types of media (TV vs. radio vs. news websites vs. social media; public service television vs. commercial television). Finally, valence of immigration coverage was coded (on a 3-point scale-positive, neutral/ambivalent, or negative) for each news item which (a) the respondents were exposed to, and (b) which contained the immigration keywords.

Control Variables
Control variables cover gender, age, income, education, interest in politics, municipality size, employment status, and region. When examining the effect of social media use, two additional variables were included: number of Facebook friends and number of Twitter accounts the respondent follows. Including these controls for variation in number and diversity of sources that the respondent would be exposed to through content shared on Facebook/Twitter. Additionally, as explained in the following, the pretest scores of each outcome variable were included in the regressions, to account for variations in attitudes prior to exposure to different campaign media.

Methods of Data Analysis
The three datasets allow for an analysis based on a pretest-posttest quasi-experimental design. Specifically, all analyses use a regressor variable method, where the dependent variable's posttest score is the outcome variable in the regression, while its pretest score is a control variable. Thus, all regressions analyze how one variable (the "intervention") predicts variation in another variable (the "posttest" scores), controlling for initial differences between groups prior to intervention (the "pretest" scores). Regressions included ordinary least squares regressions (immigration index dependent variable), binary logistic (populist vote dependent variables) and ordinal logistic regressions (ANO/SPD preference change dependent variables). All analyses and calculations are weighted by demographic distribution, and-where applicable-for sufficient adMeter participation, where participants with insufficient measurement were excluded from the analyses.

Results
The data from the first wave of our survey (pre-election) confirm that immigration has been very high on the Czech public agenda-the respondents selected it as the second most important issue facing the EU today, with an average score of 8.6 on the Likert 11-point scale. Further underscoring the perceived concerns about migration is the fact that the only issue that received a higher average score was terrorism (8.7), which is often conflated with immigration (especially from Muslim countries) in the Czech public and media discourse. 5 Nevertheless, comparing the results from the two survey waves for our key dependent variable -attitudes toward immigration-it is apparent that there was little change in views; indeed, the pre-election attitudes were on average slightly more negative (22.7) than the post-election ones (22.5), and that is the case for two out of the four individual items as well (see Figure 1). Overall, however, the surveys document that the Czech public hold strongly negative views of immigrants and are most concerned about the economic impact of immigration, followed by perceived threats to culture and the rise of criminality as a consequence of immigration.
The analysis of data with regards to our first research question-namely, how respondents' views of immigration and their electoral behavior relates to exposure to specific media types during the campaign (RQ1.1)-reveals that respondents' anti-immigration attitudes remained unaffected by the intensity of use of either of the four technologically different media types in question (see Figure 2); however, exposure to television had a small but statistically significant positive effect on voting for a populist party (B = 0.01, p < .05). In contrast, exposure to radio news had a negative impact on the likelihood to vote for the radical right-wing populist party SPD (B = −0.06, p < .05), while the use of social media also lowered the chances of voting for ANO (B = −0.02, p < .05), a finding somewhat counterintuitive given the party's reputation as masters of social media campaigning. Exploring the impact of the diversity of sources (RQ1.2), the data show that the more diverse respondents' media diet has been, the more negative their attitudes toward migrants became. The magnitude of the impact was, however, rather small; for every additional media source exposed to during the campaign, respondents' anti-immigration attitudes increased after the campaign by 0.072 points on a 25-point scale (where higher score indicated stronger anti-immigrant attitudes). With regards to the diversity of sources by technologically different media types, the same effect was shown by exposure to news websites (but not for other technologies; i.e., TV, radio, and social media), as the higher number of websites not only increased anti-immigration attitudes (B = 0.09, p < .05) but also-and much more decisively-electoral preference for the SPD party (B = 0.18, p < .05). In light of these figures, we have to reject our hypothesis H1, which was based on the assumption that a more diverse news diet would decrease chances of citizens getting trapped in echo chambers and hence decrease their anti-immigration attitudes and support for populist parties. Our findings suggest that even more diverse media diets can still stimulate antiimmigration views and attract people to right-wing populist actors in the election context. Our next research question (RQ1.3) enquired about the difference between exposure to public service and commercial television and the subsequent impact on anti-immigration attitudes and populist support. As demonstrated by Figure 2, the regression models confirm our expectations-for every additional hour exposed to media from PSM during the campaign, respondents' anti-immigration attitude decreased after the campaign by 0.013 points on a 25-point scale. On the contrary, watching commercial television was positively associated with voting for a populist party (B = 0.01, p < .01), while increasing the respondent's preferences for SPD (B = 0.02, p < .01), even if not for ANO. These findings support our hypothesis H2 and add to the growing evidence that commercial television plays an important role in nurturing voters' support for populist parties.
In the second part of our analysis, we investigate the effect of exposure to specific news stories that concerned the issue of migration. The findings, as summarized by Figure 3, indicate that while overall migration-related exposure did not change respondents' immigration attitudes (H3.1 rejected), there was a strong positive influence of migration-related content when it comes solely to news websites; those online consumers who were exposed to such content via news websites over the campaign period strengthened their antiimmigration attitudes by 1.57 points on the 25-point scale (p < .05). While no effect on attitudes was observed for exposure to broadcasting media (TV and radio), this is not the case when measuring the impact on electoral behavior. Voting for a populist party is significantly predicted by the length of exposure to migration content across different media technologies (B = 0.88, p < .05), and that is even more significant for the preferences for SPD (B = 1.76, p < .05). More hours spent watching news about migration on television also increases the likelihood of populist voting (B = 1.24, p < .05); however, the same is not true for radio listening, with overall exposure to migration content on the radio actually decreasing the likelihood of voting for a populist party (B = −1.13, p < .01). Compared to both broadcasting media, however, news websites have a much larger Migration content: all (exposure in hrs) Migration content: all (exposed or not) Migration content: TV (hrs) Migration content: TV (exposed or not) Migration content: radio (hrs) Migration content: radio (exposed or not) impact, showing that longer exposure to news online increases the chances of voting for a populist party in the 2019 EP elections-though the exact size of the effect is difficult to predict due to the large error of the estimate (the binary version of the same variable then suggests that such exposure more often leads to changing voting preferences in favor of SPD). Our H3.2 hypothesis is therefore supported, even if only partly.
In the final batch of regression models (see Figure 4), we have examined the impact of migration news content as it appeared on television channels that belong to either public service media (Czech Television ČT1 and ČT24) or the two main commercial television groups, TV Nova and TV Prima. The first model, measuring the effect on immigration attitudes, confirms the finding from previously reported analysis about public service broadcasting: we found out that the longer the exposure to migration content on public service TV, the more positive will be the attitudes toward immigration (B = −4.55, p < .01), and the binary version of the variable supports these findings too (B = −1.04, p < .01). On the contrary, exposure to migration stories aired by TV Nova (the biggest commercial TV station in the country) as well as by TV Prima (the second largest commercial channel) increases anti-immigration attitudes (H3.4 confirmed), with TV Prima having relatively much higher attitudinal impact than its main commercial rival. Watching any amount of migration content on TV Nova has, nevertheless, a positive influence on voting for a populist party (B = 0.82, p < .01), while the same effect has not been observed for TV Prima; H3.3 is therefore confirmed only partly.
The results from the analysis of valence of immigration news are broadly consistent with the findings on the exposure to immigration news in general (see Table 1). While the intensity of exposure to negative immigration news (i.e., those presenting immigration or   immigrants in a negative light) does not yield statistically significant effects on people's immigration attitudes (H3.5 rejected), it is linked with a higher likelihood of voting for a populist party (B = 1.97, p < .05), confirming thereby our hypothesis H3.6. 6 Looking at the differences between public service and commercial TVs (see Appendix C), the direction of effects of valence is mostly in line with the previously reported findings concerning the exposure to immigration news, even though most of them are not statistically significant; among the few exceptions is the finding that negative news on commercial TVs increases preferences for SPD (B = 13.72, p < .01), while news coded as neutral decreases them (B = −14.73, p < .05). However, while the p-values might be affected by the low N (only 76 items for TV news), the observed basic differences in valence of immigration news aired on Czech TV (only 17% negative) and the two commercial stations (54% negative) lend strong support for the above presented results, according to which commercial TV coverage of immigration intensifies negative attitudes toward immigrants and can support populist votes.

Discussion and Conclusions
The main objective of this study was to analyze the relationship between news consumption and attitudes toward immigration, as well as electoral support for anti-immigration populist parties, in the context of the 2019 European Parliament elections campaign in the Czech Republic. The stark contrast between the low prevalence of immigrants in the country, particularly those from the 2015 European refugee crisis, and the fact that Czechs belong to the nations with the most unfavorable attitudes towards immigrants in the EU, 7 creates fruitful research conditions for exploring the impact of mediated representation of immigration on voters' perception of the issue and their party preferences, particularly during an election campaign in which immigration was a focal point for the two main populist parties, together currently holding 50% of the seats in the Chamber of Deputies. Based on a novel combination of two-wave panel survey and digital data tracking of media exposure, the results suggest that-controlling for age, gender, education, income, region, interest in politics, and prior attitudes-certain types of media diet can have an effect on the attitudes toward immigrants, as well as on the electoral choices and affinity toward populist parties. First, the intensity of general exposure to different media types-TV, radio, news websites, or social media-did not seem to make a difference when it comes to changes in anti-immigration attitudes over the course of the election campaign; however, higher exposure to television proved to be related to the higher likelihood of voting for populist parties. Further enquiry revealed that it is not just any TV, but specifically commercial television that is contributing to the populist vote (and particularly to preferences for the far-right anti-immigration Freedom and Direct Democracy party, SPD), which is in line with results of other research studying the link between commercial television and populism (Durante et al., 2019;Schulz, 2019). On the other hand, consuming more public service television over the course of the campaign was linked to less negative attitudes to immigration at the end of the campaign. Similarly, heavier radio consumption was also negatively related to the changed preferences for SPD, meaning that the more time people spent listening to the news on the radio, the less they ended up voting for the party.
These findings are further corroborated by the more granular data on the effects of exposure to actual news about migration. Consuming migration-related news-across all media technologies, but particularly through news websites and through television-was associated with a higher chance of voting for populist parties, consistent with outcomes of several previous studies emphasizing the role of sheer salience of immigration news in growing electoral support of right-wing parties (Burscher et al., 2015;Walgrave & De Swert, 2004). Again, we have observed significant differences between public service media and commercial broadcasters; consumption of migration news from PSM television (which were predominantly neutral, as our analysis of valence has demonstrated) was linked with reduced anti-immigration prejudices at the end of the campaign, while watching the two biggest commercial TV networks, TV Nova and TV Prima (airing substantially higher proportion of negative immigration news), was linked with the hardening of the anti-immigration stances of their viewers, supporting thereby the findings from Belgium and the Netherlands that have shown a correlation between attitudes to migration and consumption of public service vs. commercial television news (Jacobs et al., 2016;Vergeer & Scheepers, 1998).
Unlike much of the findings concerning the impact of salience and tonality of immigration news, our data about the effect of diversity of media sources on anti-immigration attitudes and support for populist parties have not met our expectations. Challenging the assumption that an increased diversity of news sources may act as an antidote to echo chambers, the diversity of outlets our respondents used during the campaign -across platforms, and online sources in particular -actually appeared to increase, rather than decrease, anti-immigration attitudes and chances of voting for a populist party. While counterintuitive on the first sight, this outcome might well be interpreted as a testimony to the high overall prevalence of negative immigration-related content across the sample, as evidenced by our analysis of valence (37% of negative news, as opposed to just 12% positive); bearing in mind the above-documented impact of immigration news salience on anti-immigration prejudices, it might simply be the case that people with a relatively broader media diet (especially online) are more exposed to negative news about immigration, fueling their existing stereotypes rather than eroding them (a possibility that ought to be taken into account in future research on this topic in other national contexts, and especially in a comparative design). Another possible explanation for this finding is that citizens who are negatively disposed toward immigration, and are more likely to vote for populist parties, consume more news-and are hence more, rather than less, likely to be exposed to a more diverse news diet-compared to those who hold more neutral or positive attitudes on the issue, and are less likely to support populist parties. This interpretation is supported by the findings of a cross-country study that showed that, contrary to expectations about news avoidance among populist citizens, stronger populist attitudes actually correlated with higher levels of news exposure in general, albeit driven primarily by heavy TV news consumption (Schulz, 2019).
These findings have important implications for current debates about measures to tackle the effects of the new information ecology and especially the assumption that such effects can be tampered by exposure to a greater diversity of news sources. Rather than championing diversity as a cure, we should therefore pay closer attention to the nature of information people are exposed to (whether diverse or not) and develop a greater understanding of how their habits of news seeking and news consumption across different sources interact with their attitudes. This includes, for instance, investigating the motivations for exposure to diverse news sources and especially motivations for exposure to sources that may be out of line with one's political attitudes. As existing research suggests, some citizens may well be turning to sources they disagree with in order to reaffirm their skepticism, or because they derive pleasure from engaging with arguments with which they disagree (e.g., Tsfati & Cappella, 2010). Unpacking the interaction between motivations, news habits, and attitudes is a challenge we intend to tackle in our future research, drawing on combined insights from quantitative methods presented in this article and in-depth analysis based on qualitative methods.
Overall, due to its complex design, as well as to the scope and unique nature of the empirical data, this study represents an innovative contribution to the growing field of research exploring the impact of citizens' multifaceted news diets on their political attitudes and voting behavior, in a quickly changing media ecology where old and new media intersect. Despite its broad methodological base and utilization of real-time, longitudinal media consumption data, as opposed to commonplace but notoriously problematic measures based upon self-reported news usage, the study is not without limitations. These include the single-country design, a short time frame attached to an election campaign (not enabling it to draw conclusions about long-lasting effects of exposure), or the fact that we used a two-wave survey (rather than three-waves, which would give us more robust support for claims about causal effects). The sampling and the data collection method, relying on the novel audience measurement technology (the adMeter app), poses limitations in its own right, especially with regards to the omission of the oldest age group (70+), as well as focusing almost exclusively on online population. This invites a more cautious interpretation of some of the results which cannot be generalized to the entire Czech population, particularly when it comes to the observed impact of exposure to public service media on the reduction of anti-immigration attitudes, which might not be valid for the oldest age cohort, given their generally more conservative attitudes. For the same reason, however, the effect of watching commercial TV on populist vote should hold despite the underrepresentation of the oldest age group.
These limitations, at the same time, provide motivation for further empirical explorations and for accompanying quantitative research strategies by qualitative methods that would bring a more in-depth understanding of the ways people use diverse media resources to navigate through the new media ecology and inform their stances toward the immigration discourse.

Notes
1. According to https://www.unhcr.org/news/stories/2015/12/56ec1ebde/2015-year-europesrefugee-crisis.html, last accessed 1 September 2019. 2. As of 31 December 2017, there were 524,142 foreigners (non-Czech citizens) living in the Czech Republic, representing approximately 5% of the total population (Czech Statistical Office, 2018). 3. The unit of analysis is 1 second. The app is switched on as a default, and the panelists are sent a reminder (or given a phone call) if the central system notices the app is not sending data on a daily basis. Limitations include the inability to collect data from news apps (and print papers), or when the respondent is using unsynchronized devices such as at work computer; also, the app might sometimes keep recording exposure when the mobile phone is in the same room as the media device, but the respondent is not. Despite these limitations, the app has been accepted by the media and advertising industry as a reliable audience measurement tool in the Czech Republic; Median's own laboratory testing, simulating real-life conditions, achieved 98.67% accuracy of measurement (Median, 2018). For more details, see http://www. median.eu/en/admeter/. 4. The items were coded by two coders, following coder training and an intercoder reliability test on 10% of the sample (Krippendorff's Alpha = 0.819). 5. As the pretest questionnaire revealed, the third most important issue facing the EU today was "differences in food quality across the EU" (8.6), followed by "corruption and tax evasion" (8.4), "economy" (8.3), and "environmental issues" (8.2). 6. The seemingly counterintuitive impact of positive news on the increase of preferences for the far-right SPD party (B = 12.59, p < 0.05) might be explained by the polarizing effect of counter-attitudinal information exposure (e.g., Taber & Lodge, 2006), reinforcing-rather than altering-preexisting opinions or voting decisions, especially if these are already strong (cf. Matthes, 2012) and when the source is perceived as hostile/biased against audiences' views, which is clearly the case here, as the majority of positive news in the sample have been aired by the Czech public service broadcaster (see Appendix C), which has been a frequent target of SPD's criticism, including for its reporting on migration. 7. According to the 469 Special Eurobarometer (European Union, 2018), 51% of Czechs have a negative perception of the immigrants on society, the fourth-highest number within the EU (only exceeded by Hungary, Bulgaria and Greece); 24% displays a mixture of both positive and negative perceptions, and only 20% hold positive perceptions.

Disclosure Statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by the Economic and Social Research Council [ES/S01019X/1].